Agreed - and it's why the "atmosphere" talk misses the point.

The fans, like the team, know the rhythm of a long season — conserving noise now, saving fire for when it matters.

GW5 Results: Human v AI PL predictions ⚽🤖

🥇 Me 12, 🥈 Gemini 11, 🥉 Claude 10, ChatGPT 8

I'm gaining further ground, and a good showing for Gemini. Claude doing enough to keep the lead though.

Totals after GW4:
🥇 Claude 53
🥈 Me 45
🥉 ChatGPT 39
🏴 Gemini 35

Exactly. Same game as last season at Etihad, just reversed.

City parked the bus this time, and they weren't a player down either.

1-1. Same game, reversed.

⚽🤖 Human v AI PL predictions: GW4 results

Excuse the tardiness!

🥇 Me 13, 🥈 Claude 12, 🥉 ChatGPT 9, Gemini 8

A win for humanity in GW4 🏆, but Claude is a seriously consistent challenger nailing 8/10 (!) results this week.

Totals after GW4:
🥇 Claude 43
🥈 Me 33
🥉 ChatGPT 31
🏴 Gemini 24

P.S. Football being a low-scoring sport and my scoring system may contribute to variances too. Though there are still differences in predicting outcomes.

# of correct results out of 30 so far:

Claude - 17
Me - 14
ChatGPT - 13
Gemini - 11

Agree it's fascinating and why I started this!

Prompting after the lineup comes out is more work for me; BBC's weekly predictions article gave me the idea to get them all at once.

For science's sake, now I'll start a new chat each week with a single prompt. If they mess up, that's on them!

Here is my current prompt:

"As an expert sports analyst, provide score predictions for all Premier League matches in the current gameweek. Use real-time data on results, team form, player injuries, and other relevant information to inform your analysis."

Not a scientific process, yet! I ask for GW predictions a day before the first kick-off.

The AIs claim to use historical data and real-time factors like form/injuries, but I found they don't always search the web for the latest news, so I have to reprompt.

So I'm adapting my process weekly.

⚽🤖 Human v AI PL predictions: GW3 results

Tough week all around!

Claude 6, ChatGPT 4, Gemini & Me 2

Totals after GW3:
🥇 Claude 31
🥈 ChatGPT 22
🥉 Me 20
🏴 Gemini 16

Claude - highly accurate
ChatGPT - high-risk/high-reward
Me - correct results, inconsistent scores
Gemini - still learning

GW2 Results: Claude 14, Me 10, Gemini 8, ChatGPT 6.

Cumulative: Claude 25, Me & ChatGPT 18, Gemini 14.

Claude good at predicting exact scores.

bsky.app/profile/eesh...

GW1 Results: ChatGPT 12, Claude 11, Me 8, Gemini 6.

First round to the machines.

bsky.app/profile/eesh...

Laid out the rules and scoring system ahead of GW1 here...

Observations from GW2:

- Like GW1, Claude again got 3 perfect scores 🔮
- Gemini had more faith in West Ham than even their fans, predicted 4-4 for West Ham v Chelsea
- All 3 LLMs predicted Man Utd will beat Fulham, basically as optimistic as their fans

sorry @poorlydrawnarsenal.com if you got a few extra notifs from me earlier 🙈 — was wrestling with embeds to get the post right

Week 2 coming up. Will the AIs learn from actual results? Will my human intuition kick in if I actually pay attention to how teams looked?

Just a casual exploration of AI + football - curious to see how this plays out over a season.

What's interesting:

- 3 of us predicted Brighton 1-1 Fulham exactly (Gemini went 2-0)
- Only the human predicted Sunderland would win (gut instinct > analysis?)
- AIs were more optimistic about goal totals
- My lazy human approach was more conservative

Week 1 Results:

🥇 ChatGPT: 12/30 pts
🥈 Claude: 11/30 pts
🥉 Me (human): 8/30 pts
🏴 Gemini: 6/30 pts

The robots won Round 1... but it's early days 🤖⚽

None of us saw Sunderland 3-0 West Ham coming though...